[REBASE & FF] Refactor Early Cpu Init by os-d · Pull Request #1915 · OpenDevicePartnership/patina

os-d · 2026-10-23T18:13:38Z

Description

This PR does some refactoring to early CPU init, particularly on x64 and as it relates to confidential compute guests.

DxeCore: Move Platform Exception Handler Registration Inside Arch Layer

This moves the registration of the platform exception handlers inside the architecture specific layer so that the arch can do additional actions afterwards, as will be done in future commits.

Remove confidential_compute Feature and Detect Instead

This removes the confidential_compute feature in favor of detecting the env when possible (only after platform exception handlers are enabled) and converging on a CC guest supported path when required.

This adds the logic to detect a TDX guest, the only currently supported CC guest and moves the aliased mapping protocol installation to be gated on that. Logging between GDT install and IDT install is now required to be disabled always, because we cannot check if this is a cc guest until the platform exception handlers have been installed.

The docs are updated and because the platform exception handler is the most important integration piece, it is moved to the top of the section.

This is not considered a breaking change because CC platforms have not adopted this yet for production scenarios as the feature is still being developed.

X64: Program APIC Virtual Wire Mode On CPU Init

X64 UEFI systems rely on APIC virtual wire mode being programmed during execution, e.g. wiring up legacy interrupts to the APIC to service them. Without this, interrupts can not be serviced.

Patina has relied on pre-DXE to enable this mode or MpDxe from patina-edk2 to do so. However, MpDxe has been removed in favor of MpServices. A platform that does not set virtual wire mode pre-DXE can fail to boot.

This PR adds setting virtual wire mode for all X64 platforms, so long as the MSR for determining which APIC mode the platform supports is supported (as it has long been). The MSR is then used to determine whether this is an XAPIC or X2APIC and programs it accordingly.

X64: Make GDT and IDT install Atomic

The IDT and GDT are interrelated structures and we should not be executing code between installing our GDT and the new IDT.

This updates the code to install them in the same asm block instead of across Rust code.

DXE: Register Exception Handlers Before IDT/VBAR Install

Currently, the IDT and VBAR get installed and then we register core/platform exception handlers. However, this leaves a gap where we don't have exception handlers programmed, which causes a problem on confidential compute guests that generate virtualization exceptions early and often, including through logging.

We have handled this currently by disabling logging from GDT install through to installing the platform exception handlers.

This commit makes this more graceful and installs the core and platform exception handlers before installing the IDT/VBAR so that there is no coverage gap. The logging disable is then dropped.

How This Was Tested

Tested on Hyper-V guest FW where removing MpDxe in favor of MP services caused a hang. Adding this booted to the OS again. Tested on a physical Intel device as well booting to the OS. Tested on an enlightened TDX Hyper-V guest. Tested on Q35 and ArmVirt.

Both platforms are X2APIC code; the XAPIC code is untested but the logic is copied from edk2 handling of XAPICs.

Integration Instructions

N/A.

patina-automation · 2026-10-23T18:13:47Z

✅ QEMU Validation Passed

All QEMU validation jobs completed successfully.

Note: Q35 is only built on Windows hosts (QEMU boot is disabled due to a QEMU vfat issue).

Workflow run: https://github.com/OpenDevicePartnership/patina/actions/runs/40583220961

Boot Time to EFI Shell

Platform	Elapsed
ArmVirt (Linux Host)	30.9s
Q35 (Linux Host)	42.9s

Dependencies

Repository	Ref
patina	`9b0bd62`
patina-dxe-core-qemu	`1295159`
patina-fw-patcher	`9b100b3`
patina-qemu firmware	`v4.1.3`
patina-qemu build script	`05d359d`

Binary Size Report

Platform	Commit	Δ Debug	Δ Release	Δ Debug (LZMA)	Δ Release (LZMA)
Q35 (x86_64)	Base:`3852340` PR:`9b0bd62`	`1.22 MB` `1.23 MB` `Δ +11.00 KB`	`974.00 KB` `976.00 KB` `Δ +2.00 KB`	`469.89 KB` `473.06 KB` `Δ +3.18 KB`	`382.91 KB` `383.50 KB` `Δ +609.00 B`
ArmVirt (aarch64)	Base:`3852340` PR:`9b0bd62`	`1.12 MB` `1.11 MB` `Δ -10.00 KB`	`883.50 KB` `884.00 KB` `Δ +512.00 B`	`384.75 KB` `383.27 KB` `Δ -1.48 KB`	`313.33 KB` `313.33 KB` `Δ -1.00 B`

Compressed sizes use LZMA-alone format at preset 6 (xz --format=lzma -6).

Shared QEMU integration: 1295159.

This comment was automatically generated by the Patina QEMU PR Validation Post workflow.

cfernald

Looks good. Once the supervisor is merged we should do the cpu reefactor and merge this apic.rs with the one from mp, but not for this PR to avoid churn that affects that PR.

codecov · 2026-10-23T18:19:21Z

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

os-d · 2026-10-27T18:17:25Z

@cfernald @vineelko @makubacki I have significantly updated this PR, please re-review

This moves the registration of the platform exception handlers inside the architecture specific layer so that the arch can do additional actions afterwards, as will be done in future commits. Signed-off-by: Oliver Smith-Denny <osde@microsoft.com>

This removes the confidential_compute feature in favor of detecting the env when possible (only after platform exception handlers are enabled) and converging on a CC guest supported path when required. This adds the logic to detect a TDX guest, the only currently supported CC guest and moves the aliased mapping protocol installation to be gated on that. Logging between GDT install and IDT install is now required to be disabled always, because we cannot check if this is a cc guest until the platform exception handlers have been installed. The docs are updated and because the platform exception handler is the most important integration piece, it is moved to the top of the section. This is not considered a breaking change because CC platforms have not adopted this yet for production scenarios as the feature is still being developed. Signed-off-by: Oliver Smith-Denny <osde@microsoft.com>

X64 UEFI systems rely on APIC virtual wire mode being programmed during execution, e.g. wiring up legacy interrupts to the APIC to service them. Without this, interrupts can not be serviced. Patina has relied on pre-DXE to enable this mode or MpDxe from patina-edk2 to do so. However, MpDxe has been removed in favor of MpServices. A platform that does not set virtual wire mode pre-DXE can fail to boot. This PR adds setting virtual wire mode for all X64 platforms, so long as the MSR for determining which APIC mode the platform supports is supported (as it has long been). The MSR is then used to determine whether this is an XAPIC or X2APIC and programs it accordingly. Signed-off-by: Oliver Smith-Denny <osde@microsoft.com>

The IDT and GDT are interrelated structures and we should not be executing code between installing our GDT and the new IDT. This updates the code to install them in the same asm block instead of across Rust code. Signed-off-by: Oliver Smith-Denny <osde@microsoft.com>

Currently, the IDT and VBAR get installed and then we register core/platform exception handlers. However, this leaves a gap where we don't have exception handlers programmed, which causes a problem on confidential compute guests that generate virtualization exceptions early and often, including through logging. We have handled this currently by disabling logging from GDT install through to installing the platform exception handlers. This commit makes this more graceful and installs the core and platform exception handlers before installing the IDT/VBAR so that there is no coverage gap. The logging disable is then dropped. Signed-off-by: Oliver Smith-Denny <osde@microsoft.com>

os-d requested review from apop5, cfernald, kuqin12, makubacki and vineelko October 23, 2026 18:13

cfernald approved these changes Oct 23, 2026

View reviewed changes

vineelko approved these changes Oct 23, 2026

View reviewed changes

makubacki reviewed Oct 23, 2026

View reviewed changes

Comment thread core/patina_internal_cpu/src/interrupts/x64/apic.rs Resolved

makubacki approved these changes Oct 23, 2026

View reviewed changes

Comment thread core/patina_internal_cpu/src/interrupts/x64/apic.rs Resolved

os-d force-pushed the virtual_wire_mode branch from d76fbb0 to fbe2d91 Compare October 27, 2026 18:08

os-d changed the title ~~X64: Program APIC Virtual Wire Mode On CPU Init~~ [REBASE & FF] Refactor Early Cpu Init Oct 27, 2026

os-d force-pushed the virtual_wire_mode branch from fbe2d91 to e8cca58 Compare October 27, 2026 18:16

cfernald self-requested a review October 27, 2026 18:22

Javagedes reviewed Oct 27, 2026

View reviewed changes

Comment thread docs/src/integrate/confidential_compute.md Resolved

Javagedes approved these changes Oct 27, 2026

View reviewed changes

os-d force-pushed the virtual_wire_mode branch from e8cca58 to d6e9b73 Compare October 27, 2026 18:29

vineelko approved these changes Oct 27, 2026

View reviewed changes

Comment thread sdk/patina/src/arch.rs Resolved

Comment thread core/patina_internal_cpu/src/interrupts/stub.rs Resolved

cfernald reviewed Oct 27, 2026

View reviewed changes

Comment thread core/patina_internal_cpu/src/gdt.rs Resolved

cfernald approved these changes Oct 27, 2026

View reviewed changes

makubacki approved these changes Oct 27, 2026

View reviewed changes

Comment thread docs/src/integrate/confidential_compute.md Resolved

Comment thread core/patina_internal_cpu/src/interrupts/x64/interrupt_manager.rs Resolved

os-d added 2 commits October 27, 2026 13:11

os-d force-pushed the virtual_wire_mode branch from d6e9b73 to d943bdd Compare October 27, 2026 20:16

os-d added 3 commits October 27, 2026 13:49

os-d force-pushed the virtual_wire_mode branch from d943bdd to 9b0bd62 Compare October 27, 2026 20:49

os-d enabled auto-merge (rebase) October 27, 2026 20:49

os-d added the semver:minor Pull requests that should increment the release minor version label Oct 27, 2026

os-d merged commit 0dcdeac into OpenDevicePartnership:main Oct 27, 2026
11 checks passed

os-d deleted the virtual_wire_mode branch October 27, 2026 21:25

Navigation Menu

Repository navigation

[REBASE & FF] Refactor Early Cpu Init - #1915

[REBASE & FF] Refactor Early Cpu Init#1915
os-d merged 5 commits into
OpenDevicePartnership:mainfrom
os-d:virtual_wire_mode

os-d commented Oct 23, 2026 •

edited

Loading

Uh oh!

patina-automation Bot commented Oct 23, 2026 •

edited

Loading

Uh oh!

cfernald left a comment

Uh oh!

codecov Bot commented Oct 23, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

os-d commented Oct 27, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Conversation

os-d commented Oct 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

DxeCore: Move Platform Exception Handler Registration Inside Arch Layer

Remove confidential_compute Feature and Detect Instead

X64: Program APIC Virtual Wire Mode On CPU Init

X64: Make GDT and IDT install Atomic

DXE: Register Exception Handlers Before IDT/VBAR Install

How This Was Tested

Integration Instructions

Uh oh!

patina-automation Bot commented Oct 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

✅ QEMU Validation Passed

Boot Time to EFI Shell

Dependencies

Binary Size Report

Uh oh!

cfernald left a comment

Choose a reason for hiding this comment

Uh oh!

codecov Bot commented Oct 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Uh oh!

Uh oh!

os-d commented Oct 27, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

os-d commented Oct 23, 2026 •

edited

Loading

patina-automation Bot commented Oct 23, 2026 •

edited

Loading

codecov Bot commented Oct 23, 2026 •

edited

Loading